SQLite - UPSERT *not* INSERT or REPLACE

Question

http://en.wikipedia.org/wiki/Upsert

Insert Update stored proc on SQL Server

Is there some clever way to do this in SQLite that I have not thought of?

Basically I want to update three out of four columns if the record exists, If it does not exists I want to INSERT the record with the default (NUL) value for the fourth column.

The ID is a primary key so there will only ever be one record to UPSERT.

(I am trying to avoid the overhead of SELECT in order to determin if I need to UPDATE or INSERT obviously)

Suggestions?

---

I cannot confirm that Syntax on the SQLite site for TABLE CREATE. I have not built a demo to test it, but It doesnt seem to be supported..

If it was, I have three columns so it would actually look like:

CREATE TABLE table1( 
    id INTEGER PRIMARY KEY ON CONFLICT REPLACE, 
    Blob1 BLOB ON CONFLICT REPLACE, 
    Blob2 BLOB ON CONFLICT REPLACE, 
    Blob3 BLOB 
);

but the first two blobs will not cause a conflict, only the ID would So I asusme Blob1 and Blob2 would not be replaced (as desired)

---

UPDATEs in SQLite when binding data are a complete transaction, meaning Each sent row to be updated requires: Prepare/Bind/Step/Finalize statements unlike the INSERT which allows the use of the reset function

The life of a statement object goes something like this:

1. Create the object using sqlite3_prepare_v2()
2. Bind values to host parameters using sqlite3_bind_ interfaces.
3. Run the SQL by calling sqlite3_step()
4. Reset the statement using sqlite3_reset() then go back to step 2 and repeat.
5. Destroy the statement object using sqlite3_finalize().

UPDATE I am guessing is slow compared to INSERT, but how does it compare to SELECT using the Primary key?

Perhaps I should use the select to read the 4th column (Blob3) and then use REPLACE to write a new record blending the original 4th Column with the new data for the first 3 columns?

Answer 1

Eric B’s answer is OK if you want to preserve just one or maybe two columns from the existing row. If you want to preserve a lot of columns, it gets too cumbersome fast.

Here’s an approach that will scale well to any amount of columns on either side. To illustrate it I will assume the following schema:

 CREATE TABLE page (
     id      INTEGER PRIMARY KEY,
     name    TEXT UNIQUE,
     title   TEXT,
     content TEXT,
     author  INTEGER NOT NULL REFERENCES user (id),
     ts      TIMESTAMP DEFAULT CURRENT_TIMESTAMP
 );

Note in particular that name is the natural key of the row – id is used only for foreign keys, so the point is for SQLite to pick the ID value itself when inserting a new row. But when updating an existing row based on its name, I want it to continue to have the old ID value (obviously!).

I achieve a true UPSERT with the following construct:

 WITH new (name, title, author) AS ( VALUES('about', 'About this site', 42) )
 INSERT OR REPLACE INTO page (id, name, title, content, author)
 SELECT old.id, new.name, new.title, old.content, new.author
 FROM new LEFT JOIN page AS old ON new.name = old.name;

The exact form of this query can vary a bit. The key is the use of INSERT SELECT with a left outer join, to join an existing row to the new values.

Here, if a row did not previously exist, old.id will be NULL and SQLite will then assign an ID automatically, but if there already was such a row, old.id will have an actual value and this will be reused. Which is exactly what I wanted.

In fact this is very flexible. Note how the ts column is completely missing on all sides – because it has a DEFAULT value, SQLite will just do the right thing in any case, so I don’t have to take care of it myself.

You can also include a column on both the new and old sides and then use e.g. COALESCE(new.content, old.content) in the outer SELECT to say “insert the new content if there was any, otherwise keep the old content” – e.g. if you are using a fixed query and are binding the new values with placeholders.

Answer 2

Following Aristotle Pagaltzis and the idea of COALESCE from Eric B’s answer, here it is an upsert option to update only few columns or insert full row if it does not exist.

In this case, imagine that title and content should be updated, keeping the other old values when existing and inserting supplied ones when name not found:

NOTE id is forced to be NULL when INSERT as it is supposed to be autoincrement. If it is just a generated primary key then COALESCE can also be used (see Aristotle Pagaltzis comment).

WITH new (id, name, title, content, author)
     AS ( VALUES(100, 'about', 'About this site', 'Whatever new content here', 42) )
INSERT OR REPLACE INTO page (id, name, title, content, author)
SELECT
     old.id, COALESCE(old.name, new.name),
     new.title, new.content,
     COALESCE(old.author, new.author)
FROM new LEFT JOIN page AS old ON new.name = old.name;

So the general rule would be, if you want to keep old values, use COALESCE, when you want to update values, use new.fieldname

Answer 3

Assuming 3 columns in the table.. ID, NAME, ROLE

---

BAD: This will insert or replace all columns with new values for ID=1:

INSERT OR REPLACE INTO Employee (id, name, role) 
  VALUES (1, 'John Foo', 'CEO');

---

BAD: This will insert or replace 2 of the columns... the NAME column will be set to NULL or the default value:

INSERT OR REPLACE INTO Employee (id, role) 
  VALUES (1, 'code monkey');

---

GOOD: This will update 2 of the columns. When ID=1 exists, the NAME will be unaffected. When ID=1 does not exist, the name will be default (NULL).

INSERT OR REPLACE INTO Employee (id, role, name) 
  VALUES (  1, 
            'code monkey',
            (SELECT name FROM Employee WHERE id = 1)
          );

This will update 2 of the columns. When ID=1 exists, the ROLE will be unaffected. When ID=1 does not exist, the role will be set to 'Benchwarmer' instead of the default value.

INSERT OR REPLACE INTO Employee (id, name, role) 
  VALUES (  1, 
            'Susan Bar',
            COALESCE((SELECT role FROM Employee WHERE id = 1), 'Benchwarmer')
          );

Answer 4

SELECT COUNT(*) FROM table1 WHERE id = 1;

if COUNT(*) = 0

INSERT INTO table1(col1, col2, cole) VALUES(var1,var2,var3);

else if COUNT(*) > 0

UPDATE table1 SET col1 = var4, col2 = var5, col3 = var6 WHERE id = 1;

Answer 5

Having just read this thread and been disappointed that it wasn't easy to just to this "UPSERT"ing, I investigated further...

You can actually do this directly and easily in SQLITE.

Instead of using: INSERT INTO

Use: INSERT OR REPLACE INTO

This does exactly what you want it to do!

Answer 6

If you are generally doing updates I would ..

1. Begin a transaction
2. Do the update
3. Check the rowcount
4. If it is 0 do the insert
5. Commit

If you are generally doing inserts I would

1. Begin a transaction
2. Try an insert
3. Check for primary key violation error
4. if we got an error do the update
5. Commit

This way you avoid the select and you are transactionally sound on Sqlite.